Anthropic’s Extrapolated Model and Naming in Ethereum Scaling

Introduction: The State of Ethereum Scalability

Ethereum has been the dominant platform for decentralized applications since its launch in 2015. As adoption grew, so did the strain on its base layer. By late 2020, during the DeFi summer, users regularly paid $50-$100 in gas fees for a simple token swap. The network was processing roughly 15 transactions per second, while demand often exceeded that by a factor of ten or more.

This congestion pushed developers to look for solutions that could increase throughput without sacrificing the security guarantees that make Ethereum useful. The result is a multi-layer architecture where the base layer handles settlement and dispute resolution, while Layer-2 networks process transactions off-chain and submit compressed proofs back to the main chain.

For readers unfamiliar with the earlier scaling debate, our post on the 2019 OpenAI GPT-2 release strategy covers a parallel situation where a team deliberately withheld technology to manage risk, a dynamic that also appears in how rollup teams gradually decentralize their sequencers.

Layer-2 Scaling: The Core Approaches

Layer-2 solutions fall into two broad categories: state channels and rollups. State channels, like the Lightning Network on Bitcoin, let two parties transact off-chain and only submit the final state to the main chain. They work well for repeated payments between known parties but break down for complex DeFi interactions involving many users and dynamic state.

Rollups are the dominant approach today. They execute transactions off-chain, batch them together, and submit a compressed representation of the batch to Ethereum. The key distinction between rollup types is how they handle the proof that the batch is valid.

Optimistic Rollups

Optimistic rollups assume that submitted batches are valid unless someone challenges them. They post the transaction data to Ethereum and wait through a challenge period (typically seven days) during which any observer can submit a fraud proof showing that a specific transaction was invalid. If a fraud proof succeeds, the batch is reverted and the challenger is rewarded.

This design is simple to implement because it does not require complex cryptographic proofs. The downside is the withdrawal delay: users who want to move assets from the rollup back to Ethereum must wait the full challenge period. Projects like Arbitrum and Optimism use this model and have attracted billions in total value locked.

ZK-Rollups

Zero-knowledge rollups generate a cryptographic proof (a zk-SNARK or zk-STARK) for each batch of transactions. This proof attests that the batch was executed correctly, and Ethereum verifies the proof directly. No challenge period is needed because the proof is mathematically guaranteed to be correct.

ZK-rollups offer faster withdrawals (minutes instead of days) and lower gas costs per transaction because less data needs to be posted on-chain. The trade-off is the computational cost of generating the proofs, which can be significant. Projects like zkSync and StarkNet use this approach, and recent advances in proof generation hardware have reduced the overhead substantially.

Comparison of Layer-2 Approaches

This table summarizes the trade-offs. ZK-rollups offer better user experience for withdrawals but require more upfront computation. Optimistic rollups are simpler to deploy but impose a waiting period. Both approaches remain under active development, and the choice often depends on the specific use case.

DeFi on Layer-2: What Changes

Decentralized finance protocols that migrated to Layer-2 see lower fees and faster confirmations, but they also face new constraints. Composability, the ability to combine multiple DeFi protocols in a single transaction, works differently on rollups because each rollup is its own execution environment. A user on Arbitrum cannot directly call a contract on zkSync without bridging assets through the base layer.

Liquidity fragmentation is the biggest practical problem. If a popular lending protocol like Aave deploys on three different rollups, the liquidity pools on each rollup are separate. A user who wants to borrow against their ETH on Arbitrum must first bridge ETH from the base layer or from another rollup, paying fees and waiting for confirmations at each step.

Several projects are building cross-rollup bridges and messaging protocols to solve this. The goal is to let users move assets and call contracts across rollups as if they were on a single chain. This is still an early-stage effort, and the security assumptions of each bridge vary significantly.

For context on how infrastructure adoption patterns evolve, our analysis of OpenAI GPT-4 and Anthropic Claude 3 adoption trends in 2025 Silicon Valley shows a similar pattern of early fragmentation followed by consolidation around dominant platforms.

Security Trade-offs in Layer-2

Each Layer-2 approach introduces its own risk profile. For optimistic rollups, the main risk is that the challenge period might be insufficient if no one is watching. In practice, professional watchers (often the rollup team itself) monitor the chain and submit fraud proofs. If the watcher goes offline, a malicious batch could go unchallenged. This is why some optimistic rollups are moving toward permissionless verification where anyone can run a full node.

For ZK-rollups, the risk is in the proof system itself. A bug in the proving circuit or the verification contract could allow invalid state transitions. The mathematical proofs are sound in theory, but their software implementations have had vulnerabilities. StarkNet and zkSync have both undergone multiple security audits and bug bounty programs to mitigate this.

Bridges between Layer-2 and Layer-1 are another attack surface. The Wormhole bridge exploit in February 2022, where $320 million was stolen, showed that bridge contracts can be vulnerable even if the underlying rollup is secure. Users should understand that moving assets through a bridge means trusting the bridge’s security model, not just the rollup’s.

Data Availability and Its Implications

Data availability is a concept that matters more on Layer-2 than it might initially seem. For a rollup to be secure, users must be able to reconstruct the full state from the data posted on Ethereum. If a rollup team withholds data, users cannot verify the state and cannot withdraw their assets. This is why Ethereum’s base layer requires rollups to publish transaction data as CALLDATA rather than storing it in a separate database.

The cost of publishing data on Ethereum is a significant portion of total rollup expenses. EIP-4844 (Proto-Danksharding), scheduled for the Cancun upgrade, introduces a dedicated data blob space that will be cheaper than CALLDATA. This should reduce rollup operating costs by an estimated 80-90% for data-heavy applications.

Data availability committees (DACs) are an alternative approach used by some validium chains, where a trusted group of nodes attests that data is available without publishing it on Ethereum. This improves scalability but reintroduces trust assumptions. Most production rollups today use Ethereum for data availability because it offers the strongest guarantees.

Practical Considerations for Developers

If you are building on Layer-2, the first decision is which rollup to target. Optimistic rollups are easier to deploy because they are EVM-equivalent: you can take an existing Solidity contract and deploy it on Arbitrum or Optimism with minimal changes. ZK-rollups have more limited EVM compatibility, though zkSync’s Era and StarkNet’s Cairo language are closing the gap.

Gas estimation is different on rollups. On Ethereum, gas costs are dominated by computation. On rollups, the cost is dominated by data publication. A contract that performs many storage writes will be expensive on a rollup because each write must be posted to Ethereum. Developers should optimize for data size: pack multiple values into a single storage slot, use events instead of storage where possible, and batch transactions.

Testing on Layer-2 requires running a local rollup node. Optimism offers the op-node and op-geth stack. For ZK-rollups, zkSync provides a local testing environment that simulates the prover. In all cases, the test setup is more complex than running a local Ethereum node because it includes the sequencer, the data availability layer, and the proof generation components.

User Experience and Adoption

From a user perspective, the biggest friction point is bridging. Moving assets from Ethereum to a rollup requires a transaction on the base layer, which costs gas. Once on the rollup, transactions are cheap, but the initial deposit can be expensive during high congestion. Some wallets now integrate bridging directly, so users see a single interface for moving between layers.

Withdrawal times remain a pain point for optimistic rollups. A user who wants to exit back to Ethereum must wait seven days unless they use a fast withdrawal service that front-liquidity in exchange for a fee. These services are available from projects like Across and Hop Protocol, but they add another trust assumption and a fee.

Adoption metrics show steady growth. Total value locked on Layer-2 reached $20 billion by early 2025, up from $5 billion a year earlier. Transaction counts on Arbitrum and Optimism regularly exceed those on Ethereum mainnet. The user base is still dominated by experienced DeFi users, but wallet integrations and simpler onboarding are bringing in new participants.

For teams planning their migration strategy, our guide on upcoming npm v12 release changes and preparation strategies shows how to plan for infrastructure transitions, a skill that applies directly to choosing and migrating to a Layer-2 platform.

Future Directions

The Layer-2 roadmap includes several improvements. EIP-4844 will reduce data costs as mentioned. Account abstraction (ERC-4337) will let users pay fees in tokens other than ETH, which is especially useful on rollups where users might not hold ETH. Native rollup-to-rollup communication via shared sequencing is being explored by projects like Polygon zkEVM and Espresso Systems.

The end state is a modular Ethereum where the base layer handles consensus and data availability, while execution happens on specialized rollups optimized for different workloads. Some rollups will prioritize low latency for gaming, others will focus on privacy, and others will optimize for high throughput in DeFi. Users will move between these rollups through standardized bridges, ideally without noticing the underlying complexity.

This vision is years away. Today, the practical advice is to start with an optimistic rollup if you want EVM compatibility now, or a ZK-rollup if you can tolerate a more constrained development environment in exchange for faster withdrawals. Either way, the transition to Layer-2 is the most important scaling development in Ethereum’s history.

Sources and References

This article was researched using a combination of primary and supplementary sources:

Supplementary References

These sources provide additional context, definitions, and background information to help clarify concepts mentioned in the primary source.

• Home \ Anthropic
• Anthropic Courses
• Anthropic launches Claude Fable 5, a safer Mythos AI model: Features, availability and how to use it
• Anthropic Says Its Latest Model Is ‘Mythos-Level,’ but With Strict Safeguards